AI 情报中心

5

Import AI 448: Measuring AI R&D Automation and War Implications

行业动态商业 1 来源 · 2026-03-13 09:45:48

Jack Clark's Import AI newsletter issue 448 explores metrics for measuring AI R&D automation and discusses implications ...

5

IBM Granite 4.0 Speech: Compact Multilingual Model for Edge Devices

模型发布多模态 1 来源 · 2026-03-13 09:45:48

IBM has released Granite 4.0 1B Speech, a compact multilingual speech model optimized for edge devices. The model brings...

6

AI Research Assesses Multi-Step Cyber Attack Capabilities of Frontier Models

论文安全 1 来源 · 2026-03-13 09:45:48

New research evaluates frontier AI models on 32-step corporate and 7-step ICS (Industrial Control Systems) attack scenar...

6

NVIDIA Agent Wins DABStep Benchmark Using Data Scientist Reasoning Approach

行业动态智能体 1 来源 · 2026-03-13 09:45:48

NVIDIA's agent achieved first place on the DABStep benchmark by building an agent that thinks like a data scientist usin...

6

HuggingFace LeRobot v0.5.0 Advances AI for Real-World Robotics

开源工具智能体 1 来源 · 2026-03-13 09:45:48

HuggingFace's LeRobot project has released version 0.5.0, advancing AI capabilities for real-world robotics applications...

6

IonRouter (YC W26): GH200-Optimized Inference Engine Achieves 588 tok/s

行业动态大模型 1 来源 · 2026-03-13 09:45:48

IonRouter, a Y Combinator Winter 2026 startup, has launched a high-throughput inference engine optimized for GH200 proce...

6

Understudy: Desktop Agent That Learns from Single Demonstrations

开源工具智能体 1 来源 · 2026-03-13 09:45:48

Understudy introduces a desktop agent capable of learning tasks from a single demonstration by recording screen activity...

6

Rudel Analytics Reveals Claude Code Usage Patterns: 26% Abandonment Rate

开源工具应用 1 来源 · 2026-03-13 09:45:48

Rudel, a Claude Code session analytics tool, analyzed 1,573 sessions and found that 26% of sessions are abandoned, with ...

7

OneCLI: Rust-Based Vault for AI Agent Credential Security

开源工具安全 1 来源 · 2026-03-13 09:45:48

OneCLI provides a secure vault for AI agents built in Rust, storing credentials and providing agents with placeholder ke...

7

SWE-bench Improvements Questioned: Do Benchmarks Reflect Real-World Merge Rates?

论文应用 1 来源 · 2026-03-13 09:45:48

A critical analysis questions whether improvements on SWE-bench translate to better real-world code merge rates in produ...

7

DeepAgents SDK from LangChain Hits 25K Stars for Production AI Agents

开源工具智能体 1 来源 · 2026-03-13 09:45:48

LangChain's DeepAgents SDK has reached 25,000 GitHub stars, establishing itself as a major framework for building produc...

7

Claude Adds Interactive Visualization Capabilities for Charts and Diagrams

产品更新多模态 1 来源 · 2026-03-13 09:45:48

Claude has introduced new visualization capabilities, enabling the AI to create interactive charts, diagrams, and data v...

7

Axe: 12MB Binary Replaces AI Frameworks with Unix-Style Composable Agents

开源工具智能体 1 来源 · 2026-03-13 09:45:48

A new lightweight tool called Axe offers a 12MB binary that replaces heavy AI frameworks with composable, Unix-style age...

7

Claude Agent SDK Launches with 20K GitHub Stars for Custom Agent Development

开源工具智能体 1 来源 · 2026-03-13 09:45:48

Anthropic has released the official Claude Agent SDK, enabling developers to build custom AI agents powered by Claude. T...

8

Anthropic Invests $100M in Claude Partner Network for Enterprise AI Adoption

行业动态商业 1 来源 · 2026-03-13 09:45:48

Anthropic has committed $100 million to establish the Claude Partner Network, a major initiative to accelerate enterpris...

8

RAG Document Poisoning: 95% Attack Success Rate Exposes Critical Security Vulnerability

论文安全 1 来源 · 2026-03-13 09:45:48

New research demonstrates a 95% success rate for document poisoning attacks against RAG (Retrieval-Augmented Generation)...

8

Kotlin Creator Launches Codespeak: A Specification Language for LLM Communication

开源工具大模型 1 来源 · 2026-03-13 09:45:48

The creator of Kotlin has unveiled Codespeak, a new programming language designed specifically for communicating with LL...

9

AI Facial Recognition Wrongful Arrest Sparks National Debate on AI Safety

行业动态安全 1 来源 · 2026-03-13 09:45:48

An innocent grandmother was jailed for months after being misidentified by AI facial recognition technology in a North D...

8

Claude Code CLI Reaches 50K Stars with Agentic Coding Capabilities

开源工具智能体 1 来源 · 2026-03-13 09:45:48

Anthropic's official Claude Code CLI tool has achieved 50,000 GitHub stars, establishing itself as a leading agentic cod...

5

Sebastian Raschka Reviews 10 Open-Weight LLM Architectures

论文大模型 1 来源 · 2026-03-13 09:45:47

Sebastian Raschka published comprehensive roundup comparing 10 open-weight LLM releases from Jan-Feb 2026 including Arce...

事件列表